Secular trends in mortality by stroke subtype in the 20th century: a retrospective analysis

Transcription

1 Secular trends in mortality by stroke subtype in the 2th century: a retrospective analysis Debbie A Lawlor, George Davey Smith, David A Leon, Jonathan AC Sterne, Shah Ebrahim Summary Background Since both cerebral infarction and coronary heart disease are caused by atherosclerosis, they would be expected to have similar secular trends in mortality. Because differential diagnosis of stroke subtype on routine death certificates is inaccurate, we aimed to estimate secular trends in cerebral infarct and haemorrhage throughout the 2th century, for England and Wales, with data from autopsy studies. Methods We calculated the ratio of cerebral infarct to cerebral haemorrhage from all available sources of autopsy data from the 2th century. These data were used to estimate the ratio of cerebral infarct to haemorrhage for every year, and hence to estimate rates of cerebral infarct and cerebral haemorrhage from the total stroke mortality rate, obtained from the UK Office for National Statistics. Findings Data about stroke subtypes from autopsies were available from 1932 to The ratio of cerebral infarct to cerebral haemorrhage increased fourfold from 5 in the 193s to 2 by the 199s; most of the increase took place between the 193s and the 197s. Estimated secular trends suggested that there was a steady fall in mortality from cerebral haemorrhage throughout the 2th century, whereas mortality from cerebral infarct increased to a peak in the 197s and then fell. Trends in estimated cerebral infarct mortality closely matched those for coronary heart disease mortality. Interpretation The closely related trends in cerebral infarct and coronary heart disease suggest common causes, but the very different trend in cerebral haemorrhage shows that its cause probably differs importantly from these conditions. Published online Nov 12, 22 Department of Social Medicine, University of Bristol, Bristol, UK (D A Lawlor MPH, Prof G Davey Smith DSc, J A C Sterne PhD, Prof S Ebrahim DM); London School of Hygiene and Tropical Medicine, London (Prof D A Leon PhD) Correspondence to: D A Lawlor, Department of Social Medicine, University of Bristol, Canynge Hall, Whiteladies Road, Bristol BS8 2PR UK ( D.A.Lawlor@bristol.ac.uk) Introduction A paradox of the epidemiology of stroke is that although many of the well-known risk factors such as hypertension, smoking, physical inactivity, and obesity are common to both stroke and coronary heart disease, discordant patterns of these two diseases in relation to geography, ethnicity, and time are consistently recorded. 1 3 Differences in secular trends (changes over time) are especially noticeable. In most western countries during the past century, total stroke mortality rates declined, with no epidemic to match that recorded, especially among men, for deaths due to coronary heart disease. 4 6 Stroke is a syndrome with various pathological conditions that are not easily distinguishable clinically. 7 The two broadest subcategories are cerebral infarct and cerebral haemorrhage. Cerebral infarct is pathologically similar to coronary heart disease, and indeed most patients who have such an infarct are likely to have widespread atherosclerosis, including in the coronary arteries. 8 Differences in secular trends between stroke subtypes are therefore a possible explanation for the discordant patterns between total stroke mortality and coronary heart disease mortality. Although clinicians have tried to differentiate between stroke subtypes, even before the widespread use of brain imaging, these subcategories are not accurately diagnosed. 7,9 Reports of secular trends in stroke mortality that are based on direct analysis of routine mortality data therefore tend to present stroke as one category (total stroke). 5 The aims of our study were, first, to use ratios of infarct to haemorrhage from autopsy studies to estimate secular trends in cerebral infarct and cerebral haemorrhage throughout the 2th century for England and Wales. Second, we examined whether the trends in cerebral infarct and coronary heart disease are similar (as we would expect from our understanding of their similar pathology). Methods Mortality data Mortality rates (number of deaths per 1 resident population), standardised for age and sex, for coronary heart disease and total stroke were calculated with data obtained from the UK Office for National Statistics for Engalnd and Wales (CD Rom: ONS 2th Century Mortality [England & Wales ]). The age range years was used, because stroke and coronary heart disease are rare in people younger than 35, and diagnostic accuracy on death certificates for these conditions is less reliable for individuals older than 74 years. 5 These data were also the source of mortality rates for broad stroke subtypes (cerebral haemorrhage, cerebral infarct, and ill defined cerebrovascular disease), according to the diagnosis of the underlying cause of death from the death certificate. Since there was no separate International Classification of Disease (ICD) code for subarachnoid haemorrhage until 195 (ICD 6) this disease was included in the broad category cerebral haemorrhage for the whole period. All THE LANCET Published online November 12,

2 rates were directly standardised to the WHO European standard population. The ICD codes used to define stroke and coronary heart disease were as recommended in a review of cause-specific mortality in the 2th century (table 1). 5 Autopsy data on stroke subtypes We searched Medline (1966 September, 21) with the following exploded Medical Subject Headings (MeSH) and free text terms, and search strategy: [stroke or cerebrovascular accident or cerebral infarction or cerebral ICD 3 (1921 3) ICD 4 ( ) ICD 5 (194 49) ICD 6 (195 57) ICD 7 ( ) ICD 8 ( ) ICD 9 ( ) Cerebral infarct 74b(1) Cerebral 82b(1) Cerebral 83bc Cerebral 332 Cerebral 332 Cerebral 293 Psychosis 29 4 Arteriosclerotic embolism embolism embolus, embolus embolus associated dementia 74b(2) Cerebral 82b(2) Cerebral thrombosis and and with 433 Occlusion and thrombosis thrombosis and thrombosis thrombosis cerebral stenosis of 91b(1) Arterio- 97 Arterio- softening 333 Spasm of 333 Spasm of athero- pre-cerebral sclerosis sclerosis 97 Arterio- cerebral cerebral sclerosis arteries with cerebro- with cerebro- sclerosis arteries arteries 432 Occlusion 434 Occlusion of vascular vascular (excluding pre-cerebral cerebral lesion lesion coronary or arteries arteries renal 433 Cerebral 435 Transient stenosis or thrombosis cerebral cerebral 434 Cerebral ischaemia haemor- embolus rhage) 435 Transient ischaemic attack 437 Generalised ischaemic cerebrovascular disease Cerebral haemorrhage 74a(1) Cerebral 82a(1) Cerebral 83a Cerebral 33 Subarachnoid 33 Subarachnoid 43 Subarachnoid 43 Subarachnoid haemorrhage haemorrhage haemorrhage haemorrhage haemorrhage haemorrhage haemorrhage 331 Cerebral 331 Cerebral 431 Cerebral 431 Intracerebral haemorrhage haemorrhage haemorrhage haemorrhage 432 Other and unspecified cerebral haemorrhage Ill defined cerebrovascular lesion 74a(2) Apoplexy 82a(2) Apoplexy 83d Hemiplegia 334 Other and ill 334 Other and ill 436 Acute but ill 342 Hemiplegia 75a Hemiplegia 82c(1) Hemiplegia and other defined vascular defined vas- defined 344 Other paralytic 75b Other forms 82c(2) Other forms paralysis of lesions affecting cular lesions cerebro- syndromes of paralysis of paralysis unstated the central affecting the vascular 436 Acute but ill 83 Cerebral 82b(3) Cerebral origin nervous system central disease defined softening softening 83e Other intra- 352 Other cerebral nervous 438 Other ill cerebrocranial paralyses system defined vascular effusions 352 Other cerebro- disease cerebral vascular 437 Other ill paralyses disease defined 344 Other cerebral cerebroparalysis vascular disease 438 Late effects of cerebrovascular disease Coronary heart disease 89 Angina 93a Acute 93a Acute 42 Arteriosclerotic 42 Arteriosclerotic 41 Acute 41 Acute pectoris myocarditis myocarditis heart disease heart disease myocardial myocardial 9(5) Fatty heart 93b(1) Fatty heart 93b Chronic 422 Other 422 Other infarction infarction 9(7) Other myocar- 93b(2) Cardio- myocarditis myocardial myocardial 411 Other acute 411 Other acute and dial disease vascular 93c(1) Cardio- degeneration degeneration and subacute subacute 91b(2) Arterio- degeneration vascular 45 General 45 General forms of forms of sclerosis 93b(3) Other degeneration arteriosclerosis arterio- ischaemic ischaemic without record myocardial 93c(2)Fatty sclerosis heart disease heart disease of cerebral degeneration myocardial 412 Chronic 412 Chronic vascular 93c Myocarditis degeneration ischaemic ischaemic lesion 94 Diseases 93c(3)Other heart disease heart disease of the coro- myocardial 413 Angina 413 Angina nary arteries degeneration pectoris pectoris 97(3) Arterio- 93d Myocarditis 414 Asymptomatic 414 Asymptomatic sclerosis 94a Disease of ischaemic ischaemic without the coronary heart disease heart disease record of arteries cerebral 94b Angina vascular pectoris lesion Table 1: International Classification of Disease (ICD) codes used to define mortality data 2 THE LANCET Published online November 12, 22

3 Deaths per 1 population Ill defined cerebrovascular disease Cerebral haemorrhage Cerebral infarct ICD 3 ICD 4 ICD 5 ICD 6 ICD 7 ICD 8 ICD Figure 1: Standardised mortality rates for cerebral infarct, cerebral haemorrhage, and unclassified stroke, based on routine mortality statistics Cerebral Cerebral Ill defined infarct haemorrhage cerebrovascular disease % 62% 2% % 45% 6% % 59% 3% % 6% 5% % 53% 6% % 38% 25% % 28% 53% % 23% 64% % 29% 6% Women and men aged years, England and Wales, Table 2: Proportion of total strokes classified as cerebral infarct, cerebral haemorrhage, and ill defined cerebrovascular disease from routine mortality data haemorrhage] and [necropsy or autopsy or postmortem]. Bibliographies of retrieved articles were examined for further references and investigators were contacted, when possible, for additional information. Data were included if numbers of cerebral infarct and cerebral haemorrhage had been established at autopsy, and if investigations took place in any area of England or Wales. For recent publications based on stroke registers of both fatal and non-fatal strokes, which included diagnoses based on brain imaging, autopsy results were obtained by contact with the investigators when necessary. As with mortality data, the category cerebral haemorrhage included subarachnoid haemorrhage for the entire time period, since distinction between the two was not reported in earlier periods. Analyses were based on numbers of cerebral haemorrhages and infarcts recorded only for individuals who had been clinically diagnosed with stroke as the underlying cause of death. Those in whom another cause of death had been diagnosed clinically but who were subsequently, at autopsy, found to have died from a stroke could not be included. This was because we used the ratios obtained from autopsy studies to estimate mortality from stroke subtypes by application of these ratios to total stroke mortality obtained from routine Office for National Statistics data. These data for total stroke mortality only included those who were clinically certified as having died of a stroke. Statistical analysis Taking the mid-year for each period for which the ratio was available, we plotted the observed log infarct to haemorrhage ratios against time. We estimated smoothed trends in log ratios over time using cubic splines with knots at the minimum, maximum, and mid points of the range. 1 This plot allowed us to predict values of the log ratio for all years in the range for which data were available. The predicted ratios of infarct to haemorrhage in each year were applied to the standardised total stroke mortality rates to obtain estimated values for cerebral infarct and cerebral haemorrhage. We plotted all graphs of secular trends on a log scale because we were interested in comparison of the proportional changes in rates with time, which is done most easily on a multiplicative scale. For example, an increase in rates from 1 to 2 over a period would be described as doubling, as would an increase from 4 to 8 over the same period. If such trends are plotted on a linear scale the second increase appears much larger than the first, though both are a doubling. If they are plotted on a log scale they appear as increases of the same magnitude. Analyses were done with Stata version 7. Period of study Data source Number with Number with Ratio infarct: cerebral infarct cerebral haemorrhage haemorrhage Manchester Royal Infirmary Manchester Royal Infirmary Three largest Manchester hospitals Manchester Royal Infirmary Three largest Manchester hospitals Three largest Manchester hospitals Three largest Manchester hospitals Three largest Manchester hospitals Three largest Manchester hospitals Ten hospitals in England and Wales Three largest Manchester hospitals hospitals in England and Wales Three largest Manchester hospitals Three largest Manchester hospitals Addenbrooke s Hosptial, Cambridge Guy s Hospital, London ,15 Oxfordshire community stroke register South London community stroke register* South London community stroke register* Manchester is in the north of England; Cambridge, Oxfordshire, and London are in the south of England. *Data from Afro-Caribbean and African populations were not included in these estimates because these groups form a larger proportion of the total population in south London (23%) than they do in the whole of England and Wales. Table 3: Ratio of cerebral infarct to cerebral haemorrhage over the twentieth century and details of studies used to estimate ratios THE LANCET Published online November 12,

4 Ratio of cerebral infarct to haemorrhage Weighted ratio of infarct to haemorrhage Smoothed estimates Figure 2: Ratio of cerebral infarct to cerebral haemorrhage, based on data from autopsy studies Role of the funding source The sponsors of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. Results Figure 1 shows secular trends in the standardised mortality rates for stroke subtypes on the basis of ICD codes for the underlying cause of death obtained from routine Office for National Statistics mortality data. Between ICD 4 and ICD 8 there were striking changes in the rates for cerebral infarction, cerebral haemorrhage, and ill defined cerebrovascular disease, probably caused by shifts in diagnostic fashion. The proportion of all stroke deaths that were coded as ill defined cerebrovascular disease changed from 2% in the early 192s to less than 1% between the 193s and early 196s, and then increased substantially to over 6% by the early 199s (table 2). These observations suggest that secular trends in stroke subtypes based on diagnoses from routine death certification are likely to be strongly affected by diagnostic fashion, and, especially, changes in the propensity of clinicians to admit uncertainty in the diagnosis of stroke subtypes. Autopsy data for stroke subtype ratios for 19 periods between 1877 and 1999 were obtained (table 3). Since no Deaths per 1 population Total stroke mortality Estimated cerebral haemorrhage mortality Estimated cerebral infarction mortality Figure 3: Standardised total stroke mortality and estimated mortality for cerebral infarct and cerebral haemorrhage Estimated deaths from cerebral infarct per 1 population Coronary heart disease mortality Estimated cerebral infarct mortality ICD 5 ICD 6 ICD 7 ICD 8 ICD Deaths from coronary heart disease per 1 population Figure 4: Estimated cerebral infarct mortality and standardised coronary heart disease mortality autopsy data were available between the median year of the first period (1889) and the median year of the next period (1932) the analysis was restricted to the period Figure 2 shows the ratio of cerebral infarct to cerebral haemorrhage obtained from these studies, together with the smoothed estimates obtained by cubic splines. There was a four-fold increase in the ratio of cerebral infarct to cerebral haemorrhage from 5 in the 193s to 2 by the 199s with most of the increase between the 193s and the 197s. Figure 3 shows secular trends in age-standardised and sex-standardised mortality rates from all strokes, together with the estimated rates for cerebral infarct and cerebral haemorrhage. Estimated mortality from cerebral haemorrhage fell steadily throughout the period, and from cerebral infarct it rose from the 193s to the early 197s and thereafter declined. Figure 4 compares secular trends in coronary heart disease mortality with trends in estimated mortality from cerebral infarct. The data are scaled on separate Y axes so that the shape of the trends can be compared. The trend in estimated mortality for cerebral infarct was very similar to that for coronary heart disease. The sudden changes in mortality trends for both coronary heart disease and estimated cerebral infarct that occurred between 194 and 195 coincided with World War II and with changes in ICD coding. Figure 5 compares secular trends in coronary heart disease mortality with trends in estimated cerebral Estimated deaths from cerebral haemorrhage per 1 population Coronary heart disease mortality Estimated cerebral haemorrhage mortality ICD 5 ICD 6 ICD 7 ICD 8 ICD Figure 5: Estimated cerebral haemorrhage mortality and standardised coronary heart disease mortality Deaths from coronary heart disease per 1 population 4 THE LANCET Published online November 12, 22

5 haemorrhage mortality. Secular trends in cerebral haemorrhage differed from those from coronary heart disease: there was a steady decline in estimated cerebral haemorrhage mortality from 1932 to Discussion Total stroke mortality in England and Wales fell from the early 194s onwards. However, our estimates of stroke subtypes, based on diagnosis by autopsy, suggest that this trend masks strikingly different patterns in cerebral infarct and cerebral haemorrhage. Secular trends in estimated cerebral infarct mortality from 1932 to 1999 were similar to those for coronary heart disease mortality, whereas cerebral haemorrhage showed a steady decrease over this period. These trends suggest clear differences in causation between cerebral infarct and cerebral haemorrhage, and have important implications for disease prevention and for studies of stroke epidemiology. Secular trends in stroke subtype based on ICD codes from routine mortality data (shown in figure 1 and table 2) are strongly affected by changes in diagnostic fashion and the propensity to admit uncertainty in the diagnosis of stroke subtypes, and so are not informative. The data from which we estimated the ratio of cerebral infarct to haemorrhage were based on the results of autopsies of stroke patients admitted to hospital. Compared with cerebral infarct, cerebral haemorrhage is generally more severe or fatal, and therefore our ratios of infarct to haemorrhage based only on autopsy data might be an underestimate of the true ratio. This shortcoming should not, however, affect the pattern of change in the ratios over time. Rates of admission to hospital for stroke might have changed over time. Although contemporary stroke registers suggest that 85 9% of stroke patients are admitted, 14,16 we could find no equivalent data from the 193s. A rise in rates of admission over the time of our analysis could (if it took place mainly in patients with cerebral infarct) explain some of the increase in the cerebral infarct to haemorrhage ratio, but is unlikely to account for the major part of this trend. We applied estimated subtype ratios to the mortality rate for all individuals with an underlying cause of death coded as stroke. Within this total rate will be a proportion of non-stroke deaths. Our estimated rates for cerebral haemorrhage and infarct, therefore, do not show the exact magnitude of mortality rates for each subtype. However, the pattern of secular trends, which is our main interest, would not be affected by misclassification of some nonstroke deaths. The pronounced changes in trends in both coronary heart disease and stroke mortality during the 194s could be due to changes in ICD codes or to an effect of World War II on certification patterns. The fall in total stroke mortality in recent years could have been affected by the use of CT scanning in stroke patients, which not only distinguishes between stroke types but also differentiates between patients with stroke and those with other diseases. In the UK, CT scans were not in widespread use 15 years ago, 17 and therefore they are unlikely to have had a major effect on the downward trend in total stroke mortality. Secular trends in coronary heart disease mortality in England and Wales suggest that the 2th century epidemic of coronary heart disease affected only men. 4 We were unable to look at secular trends in stroke subtypes by sex because only one of the studies presenting autopsy data on stroke subtypes showed data separately for women and men. 14 Data from the Oxfordshire community stroke project suggest that the ratio of cerebral infarct to haemorrhage is greater for men than for women. 14 Although data are not presented separately for women and men in Yates autopsy studies, 11 he commented that rates of cerebral infarct were higher for men than for women in the 195s. If the increase in the ratio over the past century was more substantial for men than for women, then the trends we have presented here for women and men combined might underestimate the true magnitude of the epidemic of cerebral infarction in men, and as with coronary heart disease the epidemic might have occurred predominantly in men. Few studies from other countries have assessed secular trends in stroke subtypes with autopsy data. Results of such a study in one Canadian hospital showed that the ratio of cerebral infarct to haemorrhage remained constant between the periods and Several workers have looked at trends in stroke subtype using routine mortality data, but unless a high proportion of all deaths are diagnosed by autopsy such findings are likely to be inaccurate. Findings from one routine mortality data study 19 covering 196 to 199 showed similar secular trends in stroke subtype in Japan to those we have estimated from autopsy data for England and Wales. In this study, trends for the period 196 to mid 197s, when rates of cerebral infarct in Japan rose and those for cerebral haemorrhage fell, are probably accurate since autopsy rates for patients admitted to hospital in Japan throughout this time were constant and greater than 6%. 2 In Sweden, mortality rates based on routine data showed a decline in cerebral haemorrhage between 1969 and 1983 but no change in mortality rates from cerebral infarct during this period. 21 Fewer than 1% of stroke patients were diagnosed with CT scanning in Sweden during the 197s, rising to 3 4% during the 198s, 22 making the accuracy of these trends uncertain. Accurate estimates of secular trends in stroke subtypes are necessary to improve aetiological insights and monitor preventive interventions. For example, early reports of the role of cholesterol did not distinguish between stroke subtypes, and reported that serum cholesterol was not an important risk factor for stroke. The steady decline in total stroke mortality between 194 and 197, when coronary heart disease mortality and consumption of dietary fat increased, was also consistent with the conclusion that serum cholesterol was not a risk factor for stroke. However, findings of studies with accurate information on stroke subtype indicate that serum cholesterol is positively associated with cerebral infarct, but inversely associated with haemorrhagic stroke. 23 The secular trends in stroke subtypes presented here lend support to a positive association between serum cholesterol and cerebral infarct. Investigators from the 194s and 195s found no association between cigarette smoking and stroke risk, 24,25 whereas more recent studies have shown, roughly, a doubling of the risk of stroke with cigarette smoking. 26 A meta-analysis of all observational studies showed a pronounced difference in the pooled relative risks for the effect of smoking on cerebral infarct (1 9) and cerebral haemorrhage ( 7). 25 The increase in the associations of smoking with total stroke (as opposed to stroke subtypes) with time probably coincides with the changes over time in the ratio of cerebral infarct to cerebral haemorrhage that we report. Our results together with observational studies of the effect of risk factors, such as cholesterol and cigarette smoking, on stroke subtypes 23,25 suggest that cerebral infarct and coronary heart disease have very similar causes. THE LANCET Published online November 12,

6 The aetiological factors that account for the differences in secular trends between cerebral haemorrhage and both cerebral infarct and coronary heart disease are less clear, but might reflect a greater importance of early-life riskfactors in the aetiology of cerebral haemorrhage than in that of cerebral infarct and coronary heart disease. Results of two prospective studies of associations between birth size and stroke subtypes showed that birthweight is strongly inversely associated with cerebral haemorrhage but not with cerebral infarct. 27,28 Other investigators also showed in a prospective study 29 that adult height, which is related in part to nutrition and growth in childhood, showed a stronger inverse association with cerebral haemorrhage than with cerebral infarct. Number of siblings, which can be considered an indicator of material resources in the childhood home, has been found to be positively associated with cerebral haemorrhage, but not with cerebral infarct, risk in a prospective study. 3 Improvements in childhood circumstances over the past two centuries could have had an important role in the steady decline in cerebral haemorrhage over the 2th century. These results resolve the paradoxical discordance between the epidemiology of stroke and that of coronary heart disease. The closely related trends in cerebral infarct and coronary heart disease suggest a common aetiology for these diseases, but the very different trend in cerebral haemorrhage shows that its aetiology must differ importantly from that of these conditions. Contributors G Davey Smith and D Leon thought of the study idea and all authors were involved in the design. G Davey Smith, S Ebrahim, D Lawlor, and D Leon identified sources of data for the ratios of cerebral infarct to cerebral haemorrhage. D Lawlor and J Sterne undertook the statistical analysis. D Lawlor wrote the first draft of the paper and all authors contributed to the final version. D Lawlor acts as guarantor for the paper. Conflict of interest statement None declared. Acknowledgments Charles Wolfe provided additional data from the London Stroke Register. Catherine Sudlow made useful comments on an earlier draft of the paper. Debbie Lawlor is funded by the UK Medical Research Council. References 1 Davey Smith G, Ben-Shlomo Y. Geographical and social class differentials in stroke mortality the influence of early-life factors: comment on papers by Maheswaran and colleagues. J Epidemiol Community Health 1997; 51: Kuller L, Reisler DM. An explanation for variations in distribution of stroke and arteriosclerotic heart disease among populations and racial groups. Am J Epidemiol 1971; 93: Reed DM. The paradox of high risk of stroke in populations with low risk of coronary heart disease. Am J Epidemiol 199; 131: Lawlor DA, Ebrahim S, Davey Smith G. Sex matters: secular and geographical trends in sex differences in coronary heart disease mortality. BMJ 21; 323: Charlton J, Murphy ME, Khaw KT, Ebrahim SB, Davey Smith G. Cardiovascular diseases. In: Charlton J, Murphy ME, eds. The Health of Adult Britain , vol 2. London: The Stationery Office, 1997; Gale CR, Martyn CN. The conundrum of time trends in stroke. J R Soc Med 1997; 9: Allen CMC. Clinical diagnosis of the acute stroke syndrome. QJM 1983; 28: Yates PO, Hutchinson EC. Cerebral infarction: the role of stenosis of the extracranial cerebral arteries, Medical Research Council special report series no 3. London: Her Majesty s Stationery Office, Heasman MA, Lipworth L. Studies on medical and population subjects no 2: accuracy of certification of cause of death. London: Her Majesty s Stationery Office, Newson R. sg151: B-splines and splines parameterized by their values at reference points on the x-axis. Stata Technical Bulletin 2; 57: Yates PO. A change in the pattern of cerebrovascular disease. Lancet 1964; 2: Registrar General. Statistical Review of England and Wales for the London: Her Majesty s Stationery Office, Mitchinson MJ. The hypotensive stroke. Lancet 198; 1: Bamford J, Sandercock P, Dennis M, et al. A prospective study of acute cerebrovascular disease in the community: the Oxfordshire Community Stroke Project : 1 methodology, demography and incident cases of first-ever stroke. J Neurol Neurosurg Psychiatry 1988; 51: Bamford J, Sandercock P, Dennis M, Burn J, Warlow C. A prospective study of acute cerebrovascular disease in the community: the Oxfordshire Community Stroke Project : 2 incidence, case fatality rates and overall outcome at one year of cerebral infarction, primary intracerebral and subarachnoid haemorrhage. J Neurol Neurosurg Psychiatry 199; 53: Stewart JA, Dundas R, Howard RS, Rudd AG, Wolfe CD. Ethnic differences in incidence of stroke: prospective study with stroke register. BMJ 1999; 318: Sandercock P, Molyneux A, Warlow C. Value of computed tomography in patients with stroke: Oxfordshire Community Stroke Project. BMJ (Clin Res Ed) 1985; 29: Anderson TW, MacKay JS. A critical reappraisal of the epidemiology of cerebrovascular disease. Lancet 1968; 1: Kodama K. Stroke trends in Japan. Ann Epidemiol 1993; 3: Sakugawa H, Saito A. Clinical autopsy evaluation. Rinsho Byori-Japanese J Clin Pathol 1999; 47: Falkeborn M, Persson I, Terent A, Bergstrom R, Lithell H, Naessen T. Long-term trends in incidence of and mortality from acute myocardial infarction and stroke in women: analyses of total first events and of deaths in the Uppsala health care region, Sweden. Epidemiology 1996; 7: Terent A. Increasing incidence of stroke among Swedish women. Stroke 1988; 19: Hart CL, Hole DJ, Davey Smith G. The relation between cholesterol and haemorrhagic or ischaemic stroke in the Renfrew/Paisley study. J Epidemiol Community Health 2; 54: Crofton E, Crofton J. Influence of smoking on mortality from various diseases in Scotland and in England and Wales: an analysis by cohorts. BMJ 1963; II: Shinton R, Beevers G. Meta-analysis of relation between cigarette smoking and stroke. BMJ 1989; 298: Shinton R. Lifelong exposures and the potential for stroke prevention: the contribution of cigarette smoking, exercise, and body fat. J Epidemiol Community Health 1997; 51: Hypponen E, Leon DA, Kenward MG, Lithell H. Prenatal growth and risk of occlusive and haemorrhagic stroke in Swedish men and women born : historical cohort study. BMJ 21; 323: Eriksson JG, Forsen T, Tuomilehto J, Osmond C, Barker DJ. Early growth, adult income, and risk of stroke. Stroke 2; 31: McCarron P, Hart CL, Hole D, Davey Smith G. The relation between adult height and haemorrhagic and ischaemic stroke in the Renfrew/Paisley study. J Epidemiol Community Health 21; 55: Hart CL, Davey Smith G. The relationship between number of siblings and adult mortality and stroke risk: 25 year follow up of men in the collaborative study. J Epidemiol Community Health (in press) 6 THE LANCET Published online November 12, 22